Question

http://prntscr.com/6ctvls

Need an idea on an algorithm that would eliminate the needless noise parts. Signals look something like the posted link. graph acquired from a data set of roughly around 100,000 points.

Answer 1

@Kainui

Answer 2

http://prntscr.com/6ctxdv
for the needless noisy parts you can see there is no pattern to the period of the osscillations

Answer 3

there is a period that is arising from the way our vocal chords are vibrating to sound different sounds

Answer 4

Sounds like something to do with high pass/low pass filters and convolution

Answer 5

yeah, i was looking into deconvolutions to get rid of the sound, but its getting too complicated xD

Answer 6

the noise parts

Answer 7

the seems like an easy way, like for the noise parts not from our vocal chords there is no pattern to the osscilation length, but once we begin to sound our words, there is the same oscillation lengths just different amplitudes

Answer 8

I mean, can't you just ignore signals with an amplitude less than a certain height?

Answer 9

yeah I thought about htat but, if you are saying words, like sssstop, the ssss part has very low amplitude

Answer 10

and it ends up getting rid of noises in between the words too, this is okay but just wondering if theres a better way to go about it

Answer 11

You need to put a delay on it.

Answer 12

Because all waves will have nodes but you don't want to throw that out the space between words I think.

Answer 13

```
if( abs(sound) > cutoffFreq)
waitFor(delayTime)
```

Like pseudocode wise that's what I'm imagining so that you don't cut out stuff idk if that helps or not lol.

Answer 14

hold on, so why are we getting rid off the highest freq at the analog stage though

Answer 15

the green graph is voltage vs time graph, I could see the freq part and remove the higher freq part after, and deconvolve to see what the wave looks like without the high freqs

Answer 16

not deconvole i mean just do inverse fft sorrt

Answer 17

I don't know I guess I made assumptions based on your graphs but they're not really labeled. So the x axis is time and y axis is voltage?

Answer 18

ya my bad i should have labelled them. Voltage vs Time yes

Answer 19

heres what one half of the freq graph looks like the other half is mirroring it
http://prntscr.com/6cu36p

Answer 20

Wait so do you have control over the circuits as well? I think you could modify the circuit to have a high pass filter. I don't really know enough I need to learn more I guess lol.

Answer 21

yes i do, i can make a high pass,lowpass and bandpass

Answer 22

will that remove the noise of the analog part?

Answer 23

do you think its a good idea to just make a bandpass for like 50-700 hz, for human voices

Answer 24

Yeah probably not, lol

Answer 25

you should code your own DFT or just get matlab!! so u can mess with signals too lol

Answer 26

I should, that sounds fun lol. Just not sure how to start. I guess I'll find a free type thing of matlab I know there are open source versions that are similar, but I don'dt know what they're called.

Answer 27

i found this guy has great videos, helped review fourier everything
https://www.youtube.com/watch?v=8spoTzUS3o8

Answer 28

Thanks

Answer 29

@Kainui The ones I'm famiilar with are Sage and Octave:
http://www.sagemath.org/
https://www.gnu.org/software/octave/
@dan815 Did you get my PM?